Research On Key Technologies Of Flexible Robot Grasping In Unstructured Environment

The flexible cracking of the robot is one of the basic skills of intelligent robots to complete complex tasks.Nowadays,the craftsmanship will now be in a non-structural scenario,and multiple objects often appear in any position,which makes the conventional template matching method are no longer applicable.This paper proposes a flexible cracking method for a robot for unstructured environments based on deep learning,and performs experimentation in actual scenarios to verify the method.First,this paper constructs the overall framework of the robot’s grab system,the framework is generated by the non-structural scene target,and the optimal grab gesture generation and execution grabs three parts.The camera calibration principle and the acquisition of the scene image and the mechanical arm’s hand-eye system are studied.The target object scratch position is established,and the coordinate conversion relationship between the pixel coordinate system of the target object and the robot base coordinate system is studied.Second,this paper studies the FASTER RCNN neural network structure.From data enhancement,feature drawing,error control,and threshold selection to improve the traditional FASTER RCNN network structure,it will eventually get a good effect of effective target scene sensation program,which significantly improve the perceived ability of small target objects,and the target detection recognition is accurate The rate is96.6%.This paper then built a senet structure-based grab posture generating algorithm,and proposes a multi-target IOU hybrid region attitude evaluation algorithm to screen the optimal grab posture under current scenarios.Using Cornell Data Set training to crawl gesture generation network,build multi-objective test data sets to evaluate network models,and the generating accuracy of the optimal grabbing frame reaches 94.1%.Finally,this paper uses the Baxter robot.Based on the ROS robot software architecture,the Baxter robot is built to capture the experimental platform.By performing motion mode and positive,inverse kinematics,the Baxter robot coordinate transition matrix when performing crawling of the mechanical arm is performed by the Baxter robot.Transforming the target object in the Baxter robotic base coordinate system into the actual movement angle of the robot’s joints,and performs multiple crawl experiments in the real scene.The experimental results show that the improved scene target sensing algorithm and the optimal grab posture generating algorithm have good performance,and the robot’s flexible cracking technology has better robustness and grab accuracy.